RuSentNE Sentiment Attitude Extraction Studies

**UPD July 28'th, 2023:"" SentiNEREL collection is now a part of the arekit-ss project and it is possible to serialize collections just with a single shiell cmd script!

UPD May 6'th, 2023: Launched arekit-eval project for AREkit resources evaluation!

UPD February 1'st, 2023: SentiNEREL collection reader is become a part of AREkit=0.23.1!

This repository represents studies related to sentiment attitude extraction, provided for sentiment relations of the NEREL-based dataset, dubbed as SentiNEREL.

The following spreadsheet represents ML-models benchmark evaluation results obtained for the sentiment attitude relation extraction:

Leaderboard Google Spreadsheet

Powered by AREkit-0.23.0 framework, based on the tutorial: Binding a custom annotated collection for Relation Extraction.

Contents

• Installation
• Download Finetuned Models
• Serialize SentiNEREL
• Frameworks
  • AREnets directory
  • OpenNRE directory
  • DeepPavlov directory
  • Hitachi-graph-based directory
• Pretrained States
• Sponsors

Installation

pip install -r dependencies.txt

NOTE: some frameworks may require extra packages.

Collection Serialization

• arekit -- follow this tutorial to perform data serialization for arenets, opennre, and deeppavlov frameworks.

Frameworks

• opennre -- based on OpenNRE toolkit (BERT-based models).
• arenets -- based on AREkit, tensorflow-based module for neural network training/finetunning/inferring.
• deeppavlov [legacy] -- based on DeepPavlov framework (BERT-based models).
• hittachi-graph-based -- provides implementation of the graph-based approaches over transformers.

Pretrained states

• OpenNRE states

Sponsors